Advancing Prediction of Antimicrobial Resistance Using Decision Tree-Based Machine Learning Approaches: A Systematic Review

Abstract

Antimicrobial resistance is a growing global health threat, since it limits the effectiveness of traditional medicines and adds to the burden on health care systems and economies across the globe. Several techniques have been developed for modelling the patterns of data in a range of data sources (e.g. Random Forest, Gradient Boosted Decision Trees and eXtreme Gradient Boosting). The research is reported in accordance with the PRISMA guidelines and systematically reviews the data extracted from seventeen peer-reviewed articles published between 2019 and 2023 and identified using comprehensive searches of the PubMed, Scopus, IEEE Xplore, ProQuest, ASME, and Cochrane Library databases. The studies included in this overview described clinical, genomic and environmental contexts and reported consistent outperformance of classical statistical models by ensembles of decision trees with predictive accuracies and AUROC values often exceeding 0.90, especially in high-dimensional genomic data sets and intensive care unit settings. Further supporting the chances for clinical adoption, SHapley Additive Explanations, a framework for model interpretability, provided more transparency into predictions outputs beyond correctness. The results of the study provide evidence of the effectiveness of decision tree methods to inform targeted antimicrobial stewardship efforts, to optimize empiric treatment, and to support early intervention programs. These models connect computational accuracy with clinical utility, providing a scalable and flexible approach to relieve the burden of antibiotic resistance in high or low resource settings.